# Can Claude edit my notes, or only read them?

Short answer: **yes, Claude can edit your notes, not just read them.** Connected to Hjarni, it can insert a line, replace a snippet, or delete a stale bullet inside an existing note. So can ChatGPT, on the same notes. That is the part most AI memory tools skip, and it is the part that matters most.

## Reading is table stakes. Writing back is the hard part.

Almost every "AI memory" product does the same thing: the model retrieves some context and pastes it into the conversation. That is reading. It is useful, and it is also the easy half.

The hard half is keeping that context true. A read-only memory can hand Claude last quarter's pricing decision, but it can never correct it. When the decision changes, the old note just sits there, and your AI keeps citing it with total confidence. Stale retrieval is worse than no retrieval, because it looks right.

Hjarni is two-way. When something changes, you can have Claude or ChatGPT make the edit then and there. The next answer, in any client, reads the corrected version.

## How a surgical edit works

The AI does not resend two thousand words to fix one line. It makes a targeted edit:

- **Insert** a new line after or before a snippet you point at, for adding a bullet under a heading.
- **Replace** a unique snippet with new text, for fixing a stale link or a changed number.
- **Delete** a snippet cleanly, by replacing it with nothing.

Every edit points at an anchor: a piece of existing text that has to match **exactly once**. If that snippet appears twice, or not at all, the call errors and your note is left untouched. There is no guessing, and no silent edit to the wrong paragraph. This is how [editing inside a note](/changelog/edit-inside-a-note) has worked since it shipped.

## What happens when you both touch the same note

Say you are editing a note on the web while Claude updates it over MCP. The second save does not quietly overwrite the first. You land on a [conflict screen](/changelog/conflict-safe-note-editing) that tells you who saved, when, and exactly what changed, with three ways out: use the saved version, pick changes line by line, or merge by hand. Your draft is held safe until you choose. Most edits never collide; when they do, you stop losing work.

## The safety net: every version is kept

Every save is versioned. You can open a note's [history and provenance](/changelog/note-history-and-provenance), see whether you or an AI made each change, and revert to any earlier version in one step. An edit you did not want is never a lost note. It is one click back.

## Why this is the whole point

An AI second brain is only as good as how current it stays. Reading gets your context into the conversation. Editing is what keeps that context worth reading a month from now. When your AI can both, the knowledge base gets a little truer every time you use it, instead of drifting quietly out of date.

---

Hjarni is a Markdown knowledge base with a built-in MCP server that Claude and ChatGPT can both read from and write to. Connect [Claude](/docs/connect-claude-mcp) or [ChatGPT](/docs/connect-chatgpt-mcp) to your workspace and give your AI a memory it can actually keep up to date.